Simple as that. We have mostly duplicate data (copies of a postgres database under different names to be exact). And
FreeBSD 11.2-RELEASE-p5
zpool get dedup
output reflects 2.8x gains there. Combined with zfs compression 1.9x this should lead to some considerable reduction in the disk space used. But this isn't counted in df -h
output. Moreover, experimentation showed that the used space is equal to only the savings achieved with zfs compression. With dedup (2.8x ratio) or without (1.0x ratio) - after appropriate deletion & recreation of all database copies have been made - the space used as shown by df -h is exactly the same. The only thing dedup really "does" is it makes the execution of DROP DATABASE sql statements much much slower - from a few seconds to a few minutes. But where are the space gains?FreeBSD 11.2-RELEASE-p5